This notebook is one in a series of many, where we explore how different data analysis strategies affect the outcome of a proteomics experiment based on isobaric labeling and mass spectrometry. Each analysis strategy or ‘workflow’ can be divided up into different components; it is recommend you read more about that in the introduction notebook.

In this notebook specifically, we investigate the effect of varying the Differential Expression testing component on the outcome of the differential expression results. The four component variants are: moderated t-test, Wilcoxon test, permutation test, and Reproducibility-Optimized Test Statistic (ROTS).

The R packages and helper scripts necessary to run this notebook are listed in the next code chunk: click the ‘Code’ button. Each code section can be expanded in a similar fashion. You can also download the entire notebook source code. 

library(stringi)
library(gridExtra)
library(dendextend)
library(kableExtra)
library(limma)
library(psych)
library(tidyverse)
library(matrixTests)
library(coin)
library(ROTS)
source('other_functions.R')
source('plotting_functions.R')

Let’s load our PSM-level data set:

data.list <- readRDS(params$input_data_p)
dat.l <- data.list$dat.l # data in long format
display_dataframe_head(dat.l)
Mixture TechRepMixture Condition BioReplicate Run Channel Protein Peptide RT Charge PTM intensity TotalIntensity Ions.Score DeltaMZ isoInterOk noNAs onehit.protein shared.peptide
Mixture2 1 1 Mixture2_1 Mixture2_1 127C P21291 [K].gFGFGQGAGALVHSE.[-] 144.9444 2 N-Term(TMT6plex) 44048.41 394657.6 65 0.00052 TRUE TRUE FALSE FALSE
Mixture2 1 0.5 Mixture2_0.5 Mixture2_1 127N P21291 [K].gFGFGQGAGALVHSE.[-] 144.9444 2 N-Term(TMT6plex) 38298.89 394657.6 65 0.00052 TRUE TRUE FALSE FALSE
Mixture2 1 0.125 Mixture2_0.125 Mixture2_1 128C P21291 [K].gFGFGQGAGALVHSE.[-] 144.9444 2 N-Term(TMT6plex) 39552.81 394657.6 65 0.00052 TRUE TRUE FALSE FALSE
Mixture2 1 0.667 Mixture2_0.667 Mixture2_1 128N P21291 [K].gFGFGQGAGALVHSE.[-] 144.9444 2 N-Term(TMT6plex) 56730.70 394657.6 65 0.00052 TRUE TRUE FALSE FALSE
Mixture2 1 0.667 Mixture2_0.667 Mixture2_1 129C P21291 [K].gFGFGQGAGALVHSE.[-] 144.9444 2 N-Term(TMT6plex) 48744.49 394657.6 65 0.00052 TRUE TRUE FALSE FALSE
Mixture2 1 1 Mixture2_1 Mixture2_1 129N P21291 [K].gFGFGQGAGALVHSE.[-] 144.9444 2 N-Term(TMT6plex) 60179.96 394657.6 65 0.00052 TRUE TRUE FALSE FALSE

After the filtering done in data_prep.R, there are 19 UPS1 proteins remaining, even though 48 were originally spiked in.

# which proteins were spiked in?
spiked.proteins <- dat.l %>% distinct(Protein) %>% filter(stri_detect(Protein, fixed='ups')) %>% pull %>% as.character
remove_factors(spiked.proteins)
##  [1] "P02787ups"   "P02788ups"   "P01133ups"   "P69905ups"   "P01008ups"   "P02741ups"   "P01375ups"   "P10636-8ups" "O00762ups"  
## [10] "P62988ups"   "P10145ups"   "P05413ups"   "P00915ups"   "P02753ups"   "P02144ups"   "P01344ups"   "P01579ups"   "P00709ups"  
## [19] "P01127ups"
tmp=dat.l %>% distinct(Protein) %>% pull %>% as.character
# protein subsampling
if (params$subsample_p>0 & params$subsample_p==floor(params$subsample_p) & params$subsample_p<=length(tmp)){
  sub.prot <- tmp[sample(1:length(tmp), size=params$subsample_p)]
  if (length(spiked.proteins)>0) sub.prot <- c(sub.prot,spiked.proteins)
  dat.l <- dat.l %>% filter(Protein %in% sub.prot)
}

Before we begin, let’s set an RNG seed for the permutation testing, store the metadata in sample.info and show some entries below. We also pick technical replicates with a dilution factor of 0.5 as the reference condition of interest. Each condition is represented by two of eight reporter Channels in each Run.

## [1] 9107
display_dataframe_head(sample.info)
Run Run.short Channel Sample Sample.short Condition Color
Mixture1_1 Mix1_1 127C Mixture1_1:127C Mix1_1:127C 0.125 black
Mixture1_1 Mix1_1 127N Mixture1_1:127N Mix1_1:127N 0.667 green
Mixture1_1 Mix1_1 128C Mixture1_1:128C Mix1_1:128C 1 red
Mixture1_1 Mix1_1 128N Mixture1_1:128N Mix1_1:128N 0.5 blue
Mixture1_1 Mix1_1 129C Mixture1_1:129C Mix1_1:129C 0.5 blue
Mixture1_1 Mix1_1 129N Mixture1_1:129N Mix1_1:129N 0.125 black 
referenceCondition
## [1] "0.5"
channelNames
## [1] "127C" "127N" "128C" "128N" "129C" "129N" "130C" "130N"

1 Unit scale component: log2 transformation of reporter ion intensities

We use the default unit scale: the log2-transformed reportion ion intensities.

dat.unit.l <- dat.l %>% mutate(response=log2(intensity)) %>% select(-intensity)
display_dataframe_head(dat.unit.l)
Mixture TechRepMixture Condition BioReplicate Run Channel Protein Peptide RT Charge PTM TotalIntensity Ions.Score DeltaMZ isoInterOk noNAs onehit.protein shared.peptide response
Mixture2 1 1 Mixture2_1 Mixture2_1 127C P21291 [K].gFGFGQGAGALVHSE.[-] 144.9444 2 N-Term(TMT6plex) 394657.6 65 0.00052 TRUE TRUE FALSE FALSE 15.42680
Mixture2 1 0.5 Mixture2_0.5 Mixture2_1 127N P21291 [K].gFGFGQGAGALVHSE.[-] 144.9444 2 N-Term(TMT6plex) 394657.6 65 0.00052 TRUE TRUE FALSE FALSE 15.22502
Mixture2 1 0.125 Mixture2_0.125 Mixture2_1 128C P21291 [K].gFGFGQGAGALVHSE.[-] 144.9444 2 N-Term(TMT6plex) 394657.6 65 0.00052 TRUE TRUE FALSE FALSE 15.27149
Mixture2 1 0.667 Mixture2_0.667 Mixture2_1 128N P21291 [K].gFGFGQGAGALVHSE.[-] 144.9444 2 N-Term(TMT6plex) 394657.6 65 0.00052 TRUE TRUE FALSE FALSE 15.79184
Mixture2 1 0.667 Mixture2_0.667 Mixture2_1 129C P21291 [K].gFGFGQGAGALVHSE.[-] 144.9444 2 N-Term(TMT6plex) 394657.6 65 0.00052 TRUE TRUE FALSE FALSE 15.57295
Mixture2 1 1 Mixture2_1 Mixture2_1 129N P21291 [K].gFGFGQGAGALVHSE.[-] 144.9444 2 N-Term(TMT6plex) 394657.6 65 0.00052 TRUE TRUE FALSE FALSE 15.87700

2 Normalization component: medianSweeping (1)

Median sweeping means subtracting from each PSM quantification value the spectrum median (i.e., the row median computed across samples/channels) and the sample median (i.e., the column median computed across features). If the unit scale is set to intensities or ratios, the multiplicative variant of this procedure is applied: subtraction is replaced by division.

Since median sweeping needs to be applied on matrix-like data, let’s switch to wide format. (Actually, this is semi-wide, since the Channel columns still have contributions form all Runs, but that’s OK because in the next step we split by Run.)

# switch to wide format
dat.unit.w <- pivot_wider(data = dat.unit.l, id_cols=-one_of(c('Condition', 'BioReplicate')), names_from=Channel, values_from=response)
display_dataframe_head(dat.unit.w)
Mixture TechRepMixture Run Protein Peptide RT Charge PTM TotalIntensity Ions.Score DeltaMZ isoInterOk noNAs onehit.protein shared.peptide 127C 127N 128C 128N 129C 129N 130C 130N
Mixture2 1 Mixture2_1 P21291 [K].gFGFGQGAGALVHSE.[-] 144.9444 2 N-Term(TMT6plex) 394657.56 65 0.00052 TRUE TRUE FALSE FALSE 15.42680 15.22502 15.27149 15.79184 15.57295 15.87700 15.71883 15.69836
Mixture2 1 Mixture2_1 P30511 [K].wAAVVVPPGEEQR.[Y] 137.1704 2 N-Term(TMT6plex) 180359.11 27 -0.00653 TRUE TRUE TRUE FALSE 14.36513 14.14964 14.27690 14.65920 14.44955 14.58652 14.54510 14.57900
Mixture2 1 Mixture2_1 P02787ups [K].sASDLTWDNLk.[G] 148.8299 3 N-Term(TMT6plex); K11(TMT6plex) 507218.12 35 -0.00005 TRUE TRUE FALSE FALSE 16.31046 15.32833 15.11530 16.27180 15.92615 16.73355 15.74796 15.42268
Mixture2 1 Mixture2_1 Q9UBQ5 [K].iDFDSVSSIMASSQ.[-] 201.8727 2 N-Term(TMT6plex) 120944.32 72 0.00119 TRUE TRUE FALSE FALSE 13.51956 13.51995 13.66172 14.36018 13.87050 14.20738 13.86685 13.83673
Mixture2 1 Mixture2_1 Q5H9R7 [R].tGQPSAPGDTSVNGPV.[-] 106.9002 2 N-Term(TMT6plex) 99098.46 23 -0.00133 TRUE TRUE FALSE FALSE 13.54513 13.17495 13.34133 13.81323 13.58510 14.02262 13.60511 13.51803
Mixture2 1 Mixture2_1 P43490 [K].nAQLNIELEAAHH.[-] 99.8437 3 N-Term(TMT6plex) 497023.17 15 0.00061 TRUE TRUE FALSE FALSE 15.58965 15.43193 15.86127 16.18651 15.86677 16.28384 15.93733 16.03418 
dat.norm.w <- emptyList(variant.names)

2.1 median sweeping (1)

Median sweeping means subtracting from each PSM quantification value the spectrum median (i.e., the row median computed across samples/channels) and the sample median (i.e., the column median computed across features). If the unit scale is set to intensities or ratios, the multiplicative variant of this procedure is applied: subtraction is replaced by division. First, let’s sweep the medians of all the rows, and do the columns later as suggested by [Herbrich at al.](https://doi.org/10.1021/pr300624g. No need to split this per Run, because each row in this semi-wide format contains only values from one Run and each median calculation is independent of the other rows.

# subtract the spectrum median log2intensity from the observed log2intensities
dat.norm.w <- dat.unit.w
dat.norm.w[,channelNames] <- median_sweep(dat.norm.w[,channelNames], 1, '-')

3 Summarization component: Median summarization

Within each Run and within each Channel, we replace multiple related observations with their median. First, for each Peptide (median of the PSM values), then for each Protein (median of the peptide values).

# normalized data
dat.norm.summ.w <- dat.norm.w %>% group_by(Run, Protein, Peptide) %>% summarise_at(.vars = channelNames, .funs = median) %>% summarise_at(.vars = channelNames, .funs = median) %>% ungroup()

Let’s also summarize the non-normalized data for comparison later on.

# non-normalized data
# group by (run,)protein,peptide then summarize twice (once on each level)
# add select() statement because summarise_at is going bananas over character columns
dat.nonnorm.summ.w <- dat.unit.w %>% group_by(Run, Protein, Peptide) %>% select(Run, Protein, Peptide, channelNames) %>% summarise_at(.vars = channelNames, .funs = median) %>% select(Run, Protein, channelNames) %>% summarise_at(.vars = channelNames, .funs = median) %>% ungroup()

4 Normalization component: medianSweeping (2)

Now that the data is on the protein level, let’s sweep all values separately per protein in the columns/samples. This is slightly different from sweeping before the summarization step because the median of medians is not the same as the grand median, but this does not introduce any bias.

# medianSweeping: in each channel, subtract median computed across all proteins within the channel
# do the above separately for each MS run
x.split <- split(dat.norm.summ.w, dat.norm.summ.w$Run)  
x.split.norm  <- lapply(x.split, function(y) {
  y[,channelNames] <- median_sweep(y[,channelNames], 2, '-')
  return(y)
})
dat.norm.summ.w <- bind_rows(x.split.norm)

5 QC plots

Before getting to the DEA section, let’s do some basic quality control and take a sneak peek at the differences between the component variants we’ve chosen. First, however, we should make the data completely wide, so that each sample gets it’s own unique column.

# make data completely wide (also across runs)
## normalized data
dat.norm.summ.w2 <- dat.norm.summ.w %>% pivot_wider(names_from = Run, values_from = all_of(channelNames), names_glue = "{Run}:{.value}")

## non-normalized data
dat.nonnorm.summ.w2 <- dat.nonnorm.summ.w %>% pivot_wider(names_from = Run, values_from = all_of(channelNames), names_glue = "{Run}:{.value}")

5.1 Boxplots

These boxplots show that the distributions are symmetrical and centered after median sweeping normalization, as desired and expected.

# use (half-)wide format
par(mfrow=c(1,2))
boxplot_w(dat.nonnorm.summ.w,sample.info, 'raw')
boxplot_w(dat.norm.summ.w, sample.info, 'normalized')

5.2 MA plots

We then make MA plots of two single samples taken from condition 0.5 and condition 0.125, measured in different MS runs (samples Mixture2_1:127C and Mixture1_2:129N, respectively). Clearly, the normalization had a strong variance-reducing effect on the fold changes.

# use wide2 format
p1 <- maplot_ils(dat.nonnorm.summ.w2, ma.onesample.num, ma.onesample.denom, scale='log', 'raw', spiked.proteins)
p2 <- maplot_ils(dat.norm.summ.w2, ma.onesample.num, ma.onesample.denom, scale='log', 'normalized', spiked.proteins)
grid.arrange(p1, p2, ncol=2)

To increase the robustness of these results, let’s make some more MA plots, but now for all samples from condition 0.5 and condition 0.125 (quantification values averaged within condition). Indeed, even the raw, unnormalized data now show less variability. It seems that by using more samples (now 8 in both the enumerator and denominator instead of just one) in the fold change calculation the rolling average is more robust and the Sum summarization data bias has been reduced (but not disappeared).

channels.num <- sample.info %>% filter(Condition==ma.allsamples.num) %>% select(Sample) %>% pull
channels.denom <- sample.info %>% filter(Condition==ma.allsamples.denom) %>% select(Sample) %>% pull
p1 <- maplot_ils(dat.nonnorm.summ.w2, channels.num, channels.denom, scale='log', 'raw', spiked.proteins)
p2 <- maplot_ils(dat.norm.summ.w2, channels.num, channels.denom, scale='log', 'normalized', spiked.proteins)
grid.arrange(p1, p2, ncol=2)

5.3 PCA plots

Now, let’s check if these multi-dimensional data contains some kind of grouping; It’s time to make PCA plots. Even though PC1 does seem to capture the conditions, providing a gradient for the dilution number, only the 0.125 condition is completely separable in the normalized data.

5.3.1 Using all proteins

par(mfrow=c(1, 2))
pcaplot_ils(dat.nonnorm.summ.w2 %>% select(-'Protein'), info=sample.info, 'raw')
pcaplot_ils(dat.norm.summ.w2 %>% select(-'Protein'), info=sample.info, 'normalized')

There are only 19 proteins supposed to be differentially expressed in this data set, which is only a very small amount in both relative (to the 4083 proteins total) and absolute (for a biological sample) terms.

5.3.2 Using spiked proteins only

Therefore, let’s see what the PCA plots look like if we were to only use the spiked proteins in the PCA. Now, the separation between different conditions has become more distinct, which suggests the experiment was carried out successfully: only conditions 0.5 and 0.667 aren’t clearly separable.

par(mfrow=c(1, 2))
pcaplot_ils(dat.nonnorm.summ.w2 %>% filter(Protein %in% spiked.proteins) %>% select(-'Protein'), info=sample.info, 'raw')
pcaplot_ils(dat.norm.summ.w2 %>% filter(Protein %in% spiked.proteins) %>% select(-'Protein'), info=sample.info, 'normalized')

Notice how for all PCA plots, the percentage of variance explained by PC1 is now much greater than when using data from all proteins. In a real situation without spiked proteins, you might plot data corresponding to the top X most differential proteins instead.

5.4 HC (hierarchical clustering) plots

The PCA plots of all proteins has a rather lower fraction of variance explained by PC1. We can confirm this using the hierarchical clustering dendrograms below: when considering the entire multidimensional space, the different conditions are not very separable at all. This is not surprising as there is little biological variation between the conditions: there are only 19 truly differential proteins, and they all (ought to) covary in exactly the same manner (i.e., their variation can be captured in one dimension).

par(mfrow=c(1, 2))
dendrogram_ils(dat.nonnorm.summ.w2 %>% select(-Protein), info=sample.info, 'raw')
dendrogram_ils(dat.norm.summ.w2 %>% select(-Protein), info=sample.info, 'normalized')

5.5 Run effect p-value plot

Our last quality check involves a measure of how well each variant was able to assist in removing the run effect. Below are the distributions of p-values from a linear model for the response variable with Run as a covariate. If the run effect was removed successfully, these p-values ought to be large. Clearly, the raw data contains a run effect, which is partially removed by the normalization.

dat <- list(dat.nonnorm.summ.l,dat.norm.summ.l)
names(dat) <- c('raw','normalized')
run_effect_plot(dat)

6 DEA component

We wish to look at the log2 fold changes of each condition w.r.t. the reference condition with dilution ratio 0.125. Since we are working with a log2 unit scale already, this means that for each protein we just look at the difference in mean observation across all channels between one condition and the reference condition. Note that this is not the same as looking at the log2 of the ratio of mean raw intensities for each condition (left hand side below), nor the mean ratio of raw intensities for each condition (right hand side below), since \(log_2 (\frac{mean(B)}{mean(A)}) \neq \frac{mean(log_2 (B))}{mean(log_2 (A))} \neq mean(\frac{log_2 (B)}{log_2 (A)})\).

In the next subsections, we use our four components variants to check whether these fold changes are significant (criterium: \(q<0.05\)). After testing, we make a correction for multiple testing using the Benjamini-Hochberg method in order to keep the FDR under control.

6.1 Moderated t-test

The moderated t-test slightly adapted from the limma package, which in use cases like ours should improve statistical power over a regular t-test. In a nutshell, this is a t-test done independently for each protein, although the variance used in the calculation of the t-statistic is moderated using some empirical Bayes estimation.

# design matrix as used in ANOVA testing.
design.matrix <- get_design_matrix(referenceCondition, sample.info)
dat.dea <- emptyList(variant.names)
this_scale <- scale.vec
d <- column_to_rownames(as.data.frame(dat.norm.summ.w2), 'Protein')
dat.dea$moderated_ttest <- moderated_ttest(dat=d, design.matrix, scale=this_scale)

For each condition, we now get the fold changes, moderated and unmoderated p-values, moderated and unmoderated q-values (BH-adjusted p-values), and some other details: see head of dataframe below.

display_dataframe_head(dat.dea$moderated_ttest)
logFC_0.125 logFC_0.667 logFC_1 t.ord_0.125 t.ord_0.667 t.ord_1 t.mod_0.125 t.mod_0.667 t.mod_1 p.ord_0.125 p.ord_0.667 p.ord_1 p.mod_0.125 p.mod_0.667 p.mod_1 q.ord_0.125 q.ord_0.667 q.ord_1 q.mod_0.125 q.mod_0.667 q.mod_1 df.r df.0 s2.0 s2 s2.post
A0AVT1 -0.0296332 -0.0190075 0.0086228 -1.2236300 -0.7848676 0.3560550 -1.1698892 -0.7503969 0.3404173 0.2312896 0.4391218 0.7244682 0.2512504 0.4588575 0.7359131 0.9569613 0.9960123 0.9898622 0.9350748 0.9805978 0.9840728 28 2.018963 0.0056242 0.0023459 0.0025664
A0FGR8 0.0121287 0.0389655 0.0247813 0.3627376 1.1653597 0.7411452 0.3596350 1.1553923 0.7348061 0.7195248 0.2537054 0.4647764 0.7216378 0.2570437 0.4681594 0.9841943 0.9960123 0.9820911 0.9795638 0.9723943 0.9650451 28 2.018963 0.0056242 0.0044720 0.0045495
A0MZ66 0.0035130 -0.0009938 -0.0183834 0.0761086 -0.0215311 -0.3982773 0.0769941 -0.0217816 -0.4029113 0.9398739 0.9829747 0.6934465 0.9391391 0.9827663 0.6898687 0.9978394 0.9975071 0.9898622 0.9964971 0.9975410 0.9825206 28 2.018963 0.0056242 0.0085220 0.0083271
A1L0T0 0.0520381 -0.0328378 -0.0500166 0.7951963 -0.5017969 -0.7643056 0.8165216 -0.5152539 -0.7848025 0.4358354 0.6212896 0.4536089 0.4229483 0.6115101 0.4409319 0.9569613 0.9975071 0.9820911 0.9350748 0.9924981 0.9650451 20 2.018963 0.0056242 0.0128474 0.0121851
A3KMH1 0.4342224 0.3206653 0.5484563 1.7510253 1.2931002 2.2116798 1.8344616 1.3547163 2.3170662 0.0952654 0.2107143 0.0387933 0.0801348 0.1892462 0.0301887 0.9569613 0.9960123 0.7960743 0.9350748 0.9540539 0.7680706 20 2.018963 0.0056242 0.1844849 0.1680848
A4D1E9 -0.0219254 -0.0731552 -0.0426655 -0.4771987 -1.5921945 -0.9285977 -0.4796669 -1.6004298 -0.9334006 0.6383938 0.1270236 0.3641701 0.6361944 0.1237544 0.3607366 0.9741019 0.9841938 0.9820911 0.9644451 0.9260254 0.9650451 20 2.018963 0.0056242 0.0063331 0.0062681

6.2 Wilcoxon test

The Wilcoxon test (Wilcoxon, F.) is a non-parametric rank-based test for comparing two groups (i.e., biological conditions). For each protein separately, the test is applied to each condition w.r.t. the reference condition 0.5.

@Piotr: uncomment the explanation of Wilcoxon fold changes if desired, otherwise explain why not 

otherConditions <- dat.l %>% distinct(Condition) %>% pull(Condition) %>% as.character %>% sort
otherConditions <- otherConditions[-match(referenceCondition, otherConditions)]
dat.dea$Wilcoxon <- wilcoxon_test(dat.norm.summ.w2, sample.info, referenceCondition, otherConditions, logFC.method='ratio')

For each condition, we now get fold change estimates, p-values and q-values (adjusted p-values): see head of dataframe below.

display_dataframe_head(dat.dea$Wilcoxon)
logFC_0.125 logFC_0.667 logFC_1 t.mod_0.125 t.mod_0.667 t.mod_1 p.mod_0.125 p.mod_0.667 p.mod_1 q.mod_0.125 q.mod_0.667 q.mod_1
A0AVT1 -0.0294565 -0.0193514 0.0089746 21 22 34 0.2786325 0.3282051 0.8784771 0.9900182 1 1
A0FGR8 0.0132333 0.0392210 0.0259210 40 45 43 0.4418026 0.1948718 0.2786325 1.0000000 1 1
A0MZ66 -0.0002205 -0.0040481 -0.0214527 27 27 25 0.6453768 0.6453768 0.5053613 1.0000000 1 1
A1L0T0 0.0501547 -0.0362998 -0.0482751 23 14 12 0.4848485 0.5887446 0.3939394 1.0000000 1 1
A3KMH1 0.3833735 0.2894750 0.5205885 28 23 29 0.1320346 0.4848485 0.0930736 0.9900182 1 1
A4D1E9 -0.0202724 -0.0716067 -0.0423592 12 10 11 0.3939394 0.2402597 0.3095238 1.0000000 1 1

6.3 Permutation test

The Fisher-Pitman permutation test (Pitman, E.J.G.) is a classical non-parametric permutation test based on the difference between group (i.e., biological condition) means. Multiple times, the columns of the quantification matrix are randomly shuffled and the resulting shuffled data set is tested, such that a null distribution of the test statistic is constructed. The final p-values are then computed by comparing the observed test statistic value with the null distribution.

dat.dea$permutation_test <- permutation_test(dat.norm.summ.l, referenceCondition, otherConditions, seed=seed, distribution='exact')

For each condition, we now get fold change estimates, p-values and q-values (adjusted p-values): see head of dataframe below.

display_dataframe_head(dat.dea$permutation_test)
logFC_0.125 logFC_0.667 logFC_1 p.mod_0.125 p.mod_0.667 p.mod_1 q.mod_0.125 q.mod_0.667 q.mod_1
A0AVT1 -0.0296332 -0.0190075 0.0086228 0.2505051 0.3560995 0.7504274 0.9744878 1.0000000 1
A0FGR8 0.0121287 0.0389655 0.0247813 0.7123543 0.1645688 0.4610723 0.9927812 0.9978006 1
A0MZ66 0.0035130 -0.0009938 -0.0183834 0.9484071 0.9839938 0.7369075 1.0000000 1.0000000 1
A1L0T0 0.0520381 -0.0328378 -0.0500166 0.4220779 0.5887446 0.5259740 0.9749035 1.0000000 1
A3KMH1 0.4342224 0.3206653 0.5484563 0.1168831 0.2770563 0.0800866 0.9744878 0.9978006 1
A4D1E9 -0.0219254 -0.0731552 -0.0426655 0.6515152 0.1428571 0.2662338 0.9749035 0.9978006 1

6.4 ROTS

The Reproducibility-Optimized Test Statistic (ROTS) (Elo et al.) introduces two additional parameters into the denominator of the classical t-test statistic, which it uses to maximize the overlap of top X differentially abundant proteins detected in bootstrap data sets, and hence also maximize the confidence of findings.

dat.dea$ROTS <- rots_test(dat.norm.summ.w2, sample.info, referenceCondition, otherConditions)

For each condition, we now get fold change estimates, p-values and q-values (adjusted p-values): see head of dataframe below.

display_dataframe_head(dat.dea$ROTS)
logFC_0.125 logFC_0.667 logFC_1 t.mod_0.125 t.mod_0.667 t.mod_1 p.mod_0.125 p.mod_0.667 p.mod_1 q.mod_0.125 q.mod_0.667 q.mod_1
A0AVT1 -0.0296332 -0.0190075 0.0086228 0.0432651 0.6287328 -0.0281468 0.4657678 0.4094595 0.7445867 0.8145382 0.8649969 0.8806595
A0FGR8 0.0121287 0.0389655 0.0247813 -0.0175126 -1.0684886 -0.0792342 0.6911294 0.1963551 0.5225639 0.8677545 0.8154446 0.8537731
A0MZ66 0.0035130 -0.0009938 -0.0183834 -0.0049373 0.0155091 0.0550291 0.8322220 0.8752458 0.6194571 0.8819602 0.8861643 0.8715106
A1L0T0 0.0520381 -0.0328378 -0.0500166 -0.0717654 0.4836778 0.1390824 0.3112688 0.5051461 0.3468240 0.8007283 0.8664733 0.8369326
A3KMH1 0.4342224 0.3206653 0.5484563 -0.4752265 -1.1284588 -0.9781106 0.0071661 0.1760225 0.0044362 0.4461838 0.8102775 0.3413148
A4D1E9 -0.0219254 -0.0731552 -0.0426655 0.0310869 1.3375056 0.1350370 0.5607446 0.1193468 0.3561250 0.8344605 0.7335991 0.8400107

7 Results comparison

Now, the most important part: let’s find out how our component variants have affected the outcome of the DEA.

7.1 Confusion matrix

A confusion matrix shows how many true and false positives/negatives each variant has given rise to. Spiked proteins that are DE are true positives, background proteins that are not DE are true negatives. We calculate this matrix for all conditions and then calculate some other informative metrics based on the confusion matrices: accuracy, sensitivity, specificity, positive predictive value and negative predictive value.

Clearly, across the board, the non-parametric methods underperform while the moderated t-test gives rise to an acceptable but not spectacular sensitivity. That said, the contrast between conditions 0.667 and 0.5 seems not large enough to yield many significant results.

cm <- conf_mat(dat.dea, 'q.mod', 0.05, spiked.proteins)
print_conf_mat(cm, referenceCondition)
0.125 vs 0.5 contrast
moderated_ttest
Wilcoxon
permutation_test
ROTS
background spiked background spiked background spiked background spiked
not_DE 4061 4 4064 19 4064 19 4064 15
DE 3 15 0 0 0 0 0 4
0.125 vs 0.5 contrast
moderated_ttest Wilcoxon permutation_test ROTS
Accuracy 0.998 0.995 0.995 0.996
Sensitivity 0.789 0.000 0.000 0.211
Specificity 0.999 1.000 1.000 1.000
PPV 0.833 NaN NaN 1.000
NPV 0.999 0.995 0.995 0.996
0.667 vs 0.5 contrast
moderated_ttest
Wilcoxon
permutation_test
ROTS
background spiked background spiked background spiked background spiked
not_DE 4064 18 4064 19 4064 19 4064 19
DE 0 1 0 0 0 0 0 0
0.667 vs 0.5 contrast
moderated_ttest Wilcoxon permutation_test ROTS
Accuracy 0.996 0.995 0.995 0.995
Sensitivity 0.053 0.000 0.000 0.000
Specificity 1.000 1.000 1.000 1.000
PPV 1.000 NaN NaN NaN
NPV 0.996 0.995 0.995 0.995
1 vs 0.5 contrast
moderated_ttest
Wilcoxon
permutation_test
ROTS
background spiked background spiked background spiked background spiked
not_DE 4060 5 4064 19 4064 19 4064 19
DE 4 14 0 0 0 0 0 0
1 vs 0.5 contrast
moderated_ttest Wilcoxon permutation_test ROTS
Accuracy 0.998 0.995 0.995 0.995
Sensitivity 0.737 0.000 0.000 0.000
Specificity 0.999 1.000 1.000 1.000
PPV 0.778 NaN NaN NaN
NPV 0.999 0.995 0.995 0.995

7.2 Correlation scatter plots

To see whether the three Summarization methods produce similar results on the detailed level of individual proteins, we make scatter plots and check the correlation between their fold changes and between their significance estimates (q-values, in our case).

For all conditions, the q-values of the different variants correlate moderately well at best, though the Wilcoxon and permutation test correlate well (\(>0.921\)). ROTS seems consistently (over-)optimistic (i.e., low q-values, even for background proteins) even compared to the moderated t-test.

scatterplot_ils(dat.dea, significance.cols, 'q-values', spiked.proteins, referenceCondition)

On the other hand, the fold changes are very well correlated for all component variants, across all conditions, though the Wilcoxon estimates to a slightly lesser extent.

scatterplot_ils(dat.dea, logFC.cols, 'log2FC', spiked.proteins, referenceCondition)

7.3 Volcano plots

The volcano plot combines information on fold changes and statistical significance. The spike-in proteins are colored blue, and immediately it is clear that their fold changes dominate the region of statistical significance, which suggests the experiment and analysis were carried out successfully. The magenta, dashed line indicates the theoretical fold change of the spike-ins.

for (i in 1:n.contrasts){
  volcanoplot_ils(dat.dea, i, spiked.proteins, referenceCondition)}

7.4 Violin plots

A good way to assess the general trend of the fold change estimates on a more ‘macroscopic’ scale is to make a violin plot. Ideally, there will be some spike-in proteins that attain the expected fold change (red dashed line) that corresponds to their condition, while most (background) protein log2 fold changes are situated around zero.

@Piotr: results missing: add all violin plots first

# plot theoretical value (horizontal lines) and violin per variant
violinplot_ils(lapply(dat.dea[c('moderated_ttest', 'Wilcoxon')], function(x) x[spiked.proteins, logFC.cols]), referenceCondition)

8 Conclusions

It seems that the nonparametric methods do not have enough statistical power, even though ROTS is (over-)optimistic, producing low q-values for many background proteins, even compared to the moderated t-test. Though the correlation between their significance estimates is a bit all over the place, all variants do agree on the fold change estimates.

9 Session information

sessionInfo()
## R version 4.0.3 (2020-10-10)
## Platform: x86_64-pc-linux-gnu (64-bit)
## Running under: Ubuntu 18.04.5 LTS
## 
## Matrix products: default
## BLAS:   /usr/lib/x86_64-linux-gnu/blas/libblas.so.3.7.1
## LAPACK: /usr/lib/x86_64-linux-gnu/lapack/liblapack.so.3.7.1
## 
## locale:
##  [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C               LC_TIME=de_BE.UTF-8        LC_COLLATE=en_US.UTF-8    
##  [5] LC_MONETARY=de_BE.UTF-8    LC_MESSAGES=en_US.UTF-8    LC_PAPER=de_BE.UTF-8       LC_NAME=C                 
##  [9] LC_ADDRESS=C               LC_TELEPHONE=C             LC_MEASUREMENT=de_BE.UTF-8 LC_IDENTIFICATION=C       
## 
## attached base packages:
## [1] stats4    parallel  stats     graphics  grDevices utils     datasets  methods   base     
## 
## other attached packages:
##  [1] ROTS_1.18.0       coin_1.4-0        survival_3.2-7    matrixTests_0.1.9 forcats_0.5.0     stringr_1.4.0     dplyr_1.0.2      
##  [8] purrr_0.3.4       readr_1.4.0       tidyr_1.1.2       tibble_3.0.4      ggplot2_3.3.2     tidyverse_1.3.0   psych_2.0.9      
## [15] limma_3.45.19     kableExtra_1.3.1  dendextend_1.14.0 gridExtra_2.3     stringi_1.5.3     rmarkdown_2.5     pacman_0.5.1     
## 
## loaded via a namespace (and not attached):
##   [1] readxl_1.3.1          backports_1.1.10      plyr_1.8.6            splines_4.0.3         BiocParallel_1.23.3  
##   [6] TH.data_1.0-10        digest_0.6.27         foreach_1.5.1         htmltools_0.5.0       viridis_0.5.1        
##  [11] fansi_0.4.1           magrittr_1.5          doParallel_1.0.16     recipes_0.1.14        modelr_0.1.8         
##  [16] gower_0.2.2           matrixStats_0.57.0    R.utils_2.10.1        sandwich_3.0-0        colorspace_1.4-1     
##  [21] signal_0.7-6          blob_1.2.1            rvest_0.3.6           haven_2.3.1           xfun_0.18            
##  [26] libcoin_1.0-7         crayon_1.3.4          jsonlite_1.7.1        impute_1.63.0         zoo_1.8-8            
##  [31] iterators_1.0.13      glue_1.4.2            gtable_0.3.0          ipred_0.9-9           zlibbioc_1.35.0      
##  [36] webshot_0.5.2         BiocGenerics_0.35.4   scales_1.1.1          mvtnorm_1.1-1         vsn_3.57.0           
##  [41] DBI_1.1.0             Rcpp_1.0.5            viridisLite_0.3.0     tmvnsim_1.0-2         preprocessCore_1.51.0
##  [46] lava_1.6.8            prodlim_2019.11.13    httr_1.4.2            modeltools_0.2-23     ellipsis_0.3.1       
##  [51] pkgconfig_2.0.3       XML_3.99-0.5          R.methodsS3_1.8.1     farver_2.0.3          nnet_7.3-14          
##  [56] dbplyr_1.4.4          utf8_1.1.4            caret_6.0-86          tidyselect_1.1.0      labeling_0.4.2       
##  [61] rlang_0.4.8           reshape2_1.4.4        munsell_0.5.0         cellranger_1.1.0      tools_4.0.3          
##  [66] cli_2.1.0             generics_0.0.2        broom_0.7.2           evaluate_0.14         mzID_1.27.0          
##  [71] yaml_2.2.1            ModelMetrics_1.2.2.2  knitr_1.30            fs_1.5.0              packrat_0.5.0        
##  [76] ncdf4_1.17            nlme_3.1-150          R.oo_1.24.0           xml2_1.3.2            compiler_4.0.3       
##  [81] rstudioapi_0.11       png_0.1-7             e1071_1.7-4           affyio_1.59.0         reprex_0.3.0         
##  [86] highr_0.8             lattice_0.20-41       ProtGenerics_1.21.0   Matrix_1.2-18         vctrs_0.3.4          
##  [91] pillar_1.4.6          lifecycle_0.2.0       BiocManager_1.30.10   MALDIquant_1.19.3     data.table_1.13.2    
##  [96] R6_2.4.1              pcaMethods_1.81.0     affy_1.67.1           IRanges_2.23.10       codetools_0.2-16     
## [101] MASS_7.3-53           assertthat_0.2.1      withr_2.3.0           mnormt_2.0.2          multcomp_1.4-15      
## [106] S4Vectors_0.27.14     mgcv_1.8-33           hms_0.5.3             grid_4.0.3            rpart_4.1-15         
## [111] timeDate_3043.102     class_7.3-17          pROC_1.16.2           Biobase_2.49.1        lubridate_1.7.9